Opinions, Context & Ideas from the TPM Editors TPM Editor's Blog

Quick Note on the Peril of Crosstabs

This didn't make sense when Republicans in denial over Obama's mounting lead did it through September. And it doesn't make sense now. The whole line of argument is based on a pretty basic misunderstanding of polling methodologies used by most quality pollsters.

Party identification is not a demographic category. It's actually closer to candidate choice, though it's different from that too. When pollsters do polls they control for things like gender, age groups and so forth because these are real things. You don't get swings in the gender or age make up of the electorate over the course of a campaign. We know how many men and women there are in the country. We know the age scale. So if you take a poll and you get 65% women, you need to adjust that to match the population. If you see a poll and it says half the respondents are over 60, that's a problem.

For most pollsters party ID is just something that comes out where it comes out after these fixed demographic categories are adjusted for.

Here's the key -- party ID is malleable, especially if you're using a likely voter screen. Not everyone is a hard partisan. It stands to reason that people who haven't made up their minds at this late date probably don't have particularly strong partisan affiliations or for that matter even particularly strong political views. If you're undecided and then decide to vote for Mitt Romney that's going to increase the likelihood that you're going to identify as a Republican. And vice versa.

Remember too, there's a likely voter screen being applied to the poll. If one side gets particularly revved up and excited (as clearly happened with Republicans after last week's debate) that's going to push more of those folks through the likely voter screen. That won't just goose that candidates numbers it will allow shift the partisan complexion of the electorate sampled in that poll.

Now, I don't want to say there can't ever be problems with partisan sampling. There can be. But you can't look at a poll you don't like, scan through the crosstabs and say "no wonder, they didn't call the right people!"